Issues & Insights - Latest Insights - Libretto

All Money & Investing Career & Upskilling Dev & Data Culture & Liberal Arts Life & Hobbies Future & Tech Issues & Insights Management

In the AI Training Data Wars, the Government Played Its Hand First

In the AI Training Data Wars, the Government Played Its Hand First

Naver's crawling lockdown, the DOJ's intervention in the OpenAI lawsuit, and the Astra safety controversy — a record of the day three stories collided

In May 2025, the US Department of Justice filed a brief arguing that AI training on copyrighted works is fair use and vital to American competitiveness — a move that, alongside Naver's crawling restrictions and safety concerns over OpenAI's Astra model, signals a new phase in the battle over who controls AI training data.

The Day Sony and Warner Sent Anthropic the Bill, the Whole AI Industry Flinched

The Day Sony and Warner Sent Anthropic the Bill, the Whole AI Industry Flinched

A copyright lawsuit over training data has, for the first time, put a major music publisher in the plaintiff's seat

In May 2025, Anthropic was the subject of two contradictory headlines within a single day. One day it walked away clean from a retaliation suit tied to the U.S. Department of Defense; the next, Sony Music and Warner Chappell Music filed suit alleging copyright infringement.

The Government Blacklisted an AI Company for Speaking Up. A Court Reversed It.

The Government Blacklisted an AI Company for Speaking Up. A Court Reversed It.

What the Anthropic ruling quietly means for how solo operators choose their tools and structure their contracts

In May 2025, a U.S. federal court ruled that barring Anthropic from government procurement over its public stance on AI safety violated the First Amendment — a decision handed down the same day OpenAI claimed to have reached AGI and a self-improving AI system was shown outside the lab for the first time. Together, the three events point to a shift in how much any business can safely depend on a single AI vendor.

Ad

AI Usage Rate and AI Acceptance Measure Different Things

AI Usage Rate and AI Acceptance Measure Different Things

What rollout rate and acceptance rate actually reveal

A content team's AI writing tool showed 87 percent usage, yet no one mentioned it in meetings — a gap between adoption and genuine acceptance that traces back to three distinct sources of resistance: tool mismatch, role anxiety, and reward disconnect.

The Day Infrastructure, Regulation, and Security All Cracked at Once

The Day Infrastructure, Regulation, and Security All Cracked at Once

One Tuesday in May 2025, three layers of the AI ecosystem shifted within the same 24 hours — here's how to read it

On the day Nvidia announced a $3.5 billion investment in MediaTek, EU regulators applied Digital Services Act obligations to OpenAI's ChatGPT for the first time. That same night, the U.S. Department of Defense announced it would adopt ChatGPT and Elon Musk's Grok as official work tools.

Why AI Falls for the Fake-Data Confidence Trap

Why AI Falls for the Fake-Data Confidence Trap

A new study finds that AI's certainty has almost nothing to do with whether the data behind it is real.

A new study finds that pairing an AI model with a market panel makes it answer far more confidently — even when every number in that panel is fabricated. The real fix isn't better data, but a single decision gate that governs whether the model should venture an answer at all.

공식 명분이 가장 요란할 때 실제 자원은 가장 조용히 이동합니다

공식 명분이 가장 요란할 때 실제 자원은 가장 조용히 이동합니다

산업 전환기의 두 시장: 공식 명분이 거래되는 곳과 인프라 접근권이 굳어지는 곳

2008년 애플 앱스토어가 문을 열었을 때, 가장 많이 인용된 말은 이것이었습니다. "개발자 누구나 전 세계 수억 명의 손안에 소프트웨어를 올릴 수 있다." 그 말은 거짓이 아니었습니다. 그런데 그 혁명의 수익 배분 구조를 설계한 것은 앱이 아니라 플랫폼이었습니다.

What Makes Readers Click Follow — and What Doesn't

What Makes Readers Click Follow — and What Doesn't

Traffic a platform distributes and traffic readers choose for themselves respond very differently to outside shocks

In August 2026, Google added a publisher follow button to Search, Discover, and Google News, letting readers mark an outlet as a preferred source so its content surfaces more often afterward — a lifeline extended to publishers just as AI search eats into web traffic.

One False Sentence Is Enough to Poison an AI Agent's Memory

One False Sentence Is Enough to Poison an AI Agent's Memory

If you've been piling client notes, vendor conversations, and purchase orders into your AI assistant's memory, a new study suggests that entire memory store can be destabilized by a surprisingly small amount of false information — and users often have no way of knowing when it happened.

Ad

When Published Books Become AI Training Data: What Copyright Protects — and What It Doesn't

When Published Books Become AI Training Data: What Copyright Protects — and What It Doesn't

Why human reading and AI data training look alike, and where the law draws the line between them

Reading a book to absorb its style and feeding hundreds of thousands of books into a writing model as training data look similar on the surface — both draw something from someone else's text and both produce something new. Yet one faces no restrictions at all, while the other is currently being litigated simultaneously in multiple US federal courts.

What AI Companies Actually Bought — and Threw Away — When They Destroyed Rare Books

What AI Companies Actually Bought — and Threw Away — When They Destroyed Rare Books

What scan-then-shred reveals about the new architecture of the knowledge trade

When Anna's Archive published a blog post this year describing how AI companies scanned and then physically destroyed rare books, Hacker News split down the middle — some called it a cultural crime, others called it sound business logic. Both reactions can be right at once, and tracing why reveals what these companies actually extracted from the books, and what they left behind.

Pick AI Tools by Writing Down the Time They'll Save

Pick AI Tools by Writing Down the Time They'll Save

Rippling didn't cancel subscriptions or slash budgets. It just made it possible to see where the money goes first.

HR platform Rippling launched a real-time AI spend-tracking tool this month, showing exactly which employees and teams pay for which AI subscriptions. The company says it built the tool after spending millions of dollars on AI tools over several months and struggling to figure out where the money had gone.

B2B 솔루션 업체의 성장 수치, 계약 전에 확인해야할 것

B2B 솔루션 업체의 성장 수치, 계약 전에 확인해야할 것

피치덱의 매출 성장률과 현금흐름표가 같은 이야기를 하지 않을 때

2025년 하반기, 식자재마트 디지털 전환 솔루션을 공급하던 한 스타트업이 서비스를 종료했습니다. 대표가 사퇴했고, 솔루션을 쓰던 업체들은 갑자기 시스템 접근이 막히거나 운영 데이터를 회수하는 과정에서 혼란을 겪었습니다. 이 회사의 피치덱에는 2023년 대비 2024년 매출 237% 성장, 2024년 대비 2025년 166% 성장이 적혀 있었습니다.

Twitch's AI Training Policy: Why Opt-In and Opt-Out Aren't Equivalent

Twitch's AI Training Policy: Why Opt-In and Opt-Out Aren't Equivalent

What Twitch's AI Policy Revealed About the Unequal Weight of Opt-In and Opt-Out

In August 2026, Twitch's chief product officer admitted live on stream that the platform's new AI-training policy relied on opt-out consent because no streamer would have agreed to it under opt-in — a slip that exposes how defaults, not choices, end up writing the terms of the deal.

The Kano Model: How Overnight Delivery Went From Delight to Baseline

The Kano Model: How Overnight Delivery Went From Delight to Baseline

What the Kano Model Reveals About Shifting Expectations

Dawn delivery, free trials, and AI recommendations once delighted users. Now their absence triggers complaints—and the Kano Model explains why yesterday's delight becomes today's baseline.

Ad

검색 유입 공식이 깨졌다, 좋은 글을 올리면 독자가 온다는 전제

검색 유입 공식이 깨졌다, 좋은 글을 올리면 독자가 온다는 전제

기존의 공식이 흔들리고 있습니다

인터넷이 대중화된 이후, 콘텐츠를 만드는 사람들이 공통으로 들어온 조언이 있습니다. 좋은 글을 꾸준히, 공개적으로 올려두면 검색 엔진이 알아서 독자를 데려온다는 것입니다. 블로그, 뉴스레터, 유튜브, 팟캐스트. 플랫폼이 바뀌어도 기저 논리는 같았습니다. 더 많이 공개할수록 더 많이 발견되고, 더 많이 발견될수록 더 많은 사람이 온다.

올버즈 시총 41억 달러의 붕괴, 브랜드 자산이 고객군에 묶여 있을 때

올버즈 시총 41억 달러의 붕괴, 브랜드 자산이 고객군에 묶여 있을 때

브랜드 자산은 특정 고객군에 묶인 자원이다

2021년 11월 나스닥 상장 당일, 올버즈의 시가총액은 41억 달러였습니다. 창업 5년 만에 양모와 사탕수수로 만든 운동화 회사가 받은 평가였습니다. 그 숫자는 신발 회사의 몸값이 아니었습니다. "소비자가 친환경 프리미엄에 돈을 더 낸다"는 명제에 붙은 가격이었습니다. 그 명제는 4년 뒤 주가 99% 하락으로 검증 결과를 받았습니다.

Subscribers Vanished in a Naver Redesign: What Creators Actually Own

Subscribers Vanished in a Naver Redesign: What Creators Actually Own

A way to measure the gap between your platform numbers and what you actually own

A beauty director spent seven years building an audience on Naver Blog — tens of thousands of followers, hundreds of thousands of monthly views. Then Naver restructured its Influencer Search tab, subscriber notifications started behaving differently, and new posts stopped reaching existing followers. The content was still sitting on the server.

What Investors Are Actually Measuring When They Listen to Your Pitch

What Investors Are Actually Measuring When They Listen to Your Pitch

The signals seed investors track instead of the idea itself

Only 15 to 20 percent of seed-funded startups go on to raise a Series A, and CB Insights data shows the dropout usually isn't about the idea — "no market need" (42%) and running out of cash (29%) top the list of failure causes. Former NEA partner Vanessa Larco's litmus test for AI pitches reveals what investors are really screening for.

What Happened When We Handed a Real Business to GPT-5.6 Sol

What Happened When We Handed a Real Business to GPT-5.6 Sol

How the system ended up sending customers false information

Posted on Hacker News, this case drew 242 upvotes and 146 comments, split between two camps — "AI shouldn't run a business" and "it's a prompt design problem." Both are right, but the more interesting question is why these outcomes were logically consistent from the AI's own point of view.

How Deep Learning Actually Works, Explained Through AlphaGo

How Deep Learning Actually Works, Explained Through AlphaGo

Most people use ChatGPT every day without knowing how it actually learns, and the same mystery surrounds AlphaGo's famous win over Lee Sedol. This piece breaks down how deep learning really works through two turning points: AlexNet and AlphaGo.

Ad

How to Start Vibe Coding: Building Apps Without Writing a Line of Code

How to Start Vibe Coding: Building Apps Without Writing a Line of Code

What to keep in mind when you start building something with vibe coding

It's no longer unusual to see someone who has never learned to code type instructions into a chatbot and end up with a working app or website. This piece traces why "vibe coding" has become possible now, where that shift began, and what you should tackle first.

What a 180-Year-Old Institution Wouldn't Sell to AI

What a 180-Year-Old Institution Wouldn't Sell to AI

The business model was open for negotiation. One thing never was.

The Associated Press has never once run an editorial opinion in 180 years. As it negotiates AI partnerships today, its CEO draws a hard line between what's negotiable and what isn't — a distinction that translates directly to solo creators and knowledge entrepreneurs facing the same choice.

The Perfect Marketing Machine Burns Out Fastest

The Perfect Marketing Machine Burns Out Fastest

A brand formula gets copied the moment it's perfected — and grows stale the instant it's copied.

In 2025, Red Bull poured roughly 30 percent of its revenue — more than $4 billion a year — into marketing a single can of energy drink, running two F1 teams and a media house that generates billions of views annually. This July, The Economist diagnosed the first real warning sign in that machine.